| Message ID | 20180706173049.e28171504c745934a280dee8@linux-foundation.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Andrew, On Fri, Jul 06, 2018 at 05:30:49PM -0700, Andrew Morton wrote: > On Tue, 19 Jun 2018 13:53:08 +0100 Will Deacon <will.deacon@arm.com> wrote: > > > In preparation for implementing the asm-generic atomic bitops in terms > > of atomic_long_*, we need to prevent asm/atomic.h implementations from > > pulling in linux/bitops.h. A common reason for this include is for the > > BITS_PER_BYTE definition, so move this and some other BIT() and masking > > macros into a new header file, linux/bits.h > > > > --- a/include/linux/bitops.h > > +++ b/include/linux/bitops.h > > @@ -2,29 +2,9 @@ > > #ifndef _LINUX_BITOPS_H > > #define _LINUX_BITOPS_H > > #include <asm/types.h> > > +#include <linux/bits.h> > > > > -#ifdef __KERNEL__ > > -#define BIT(nr) (1UL << (nr)) > > -#define BIT_ULL(nr) (1ULL << (nr)) > > -#define BIT_MASK(nr) (1UL << ((nr) % BITS_PER_LONG)) > > -#define BIT_WORD(nr) ((nr) / BITS_PER_LONG) > > -#define BIT_ULL_MASK(nr) (1ULL << ((nr) % BITS_PER_LONG_LONG)) > > -#define BIT_ULL_WORD(nr) ((nr) / BITS_PER_LONG_LONG) > > -#define BITS_PER_BYTE 8 > > #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(long)) > > -#endif > > Why does it leave BITS_TO_LONGS() in place? > > That becomes unfortunate with Chris's patch, so I'm moving > BITS_TO_LONGS() into bits.h. The reason I avoided that was because it would pull in the dreaded kernel.h for DIV_ROUND_UP, and then we're back to circular include hell :( Will > From: Chris Wilson <chris@chris-wilson.co.uk> > Subject: include/linux/bitops.h: introduce BITS_PER_TYPE > > net_dim.h has a rather useful extension to BITS_PER_BYTE to compute the > number of bits in a type (BITS_PER_BYTE * sizeof(T)), so promote the macro > to bitops.h, alongside BITS_PER_BYTE, for wider usage. > > Link: http://lkml.kernel.org/r/20180706094458.14116-1-chris@chris-wilson.co.uk > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Reviewed-by: Jani Nikula <jani.nikula@intel.com> > Cc: Randy Dunlap <rdunlap@infradead.org> > Cc: Andy Gospodarek <gospo@broadcom.com> > Cc: David S. Miller <davem@davemloft.net> > Cc: Thomas Gleixner <tglx@linutronix.de> > Cc: Ingo Molnar <mingo@kernel.org> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org> > --- > > include/linux/bitops.h | 3 ++- > include/linux/net_dim.h | 1 - > 2 files changed, 2 insertions(+), 2 deletions(-) > > diff -puN include/linux/bitops.h~bitops-introduce-bits_per_type include/linux/bitops.h > --- a/include/linux/bitops.h~bitops-introduce-bits_per_type > +++ a/include/linux/bitops.h > @@ -11,7 +11,8 @@ > #define BIT_ULL_MASK(nr) (1ULL << ((nr) % BITS_PER_LONG_LONG)) > #define BIT_ULL_WORD(nr) ((nr) / BITS_PER_LONG_LONG) > #define BITS_PER_BYTE 8 > -#define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(long)) > +#define BITS_PER_TYPE(type) (sizeof(type) * BITS_PER_BYTE) > +#define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_TYPE(long)) > #endif > > /* > diff -puN include/linux/net_dim.h~bitops-introduce-bits_per_type include/linux/net_dim.h > --- a/include/linux/net_dim.h~bitops-introduce-bits_per_type > +++ a/include/linux/net_dim.h > @@ -363,7 +363,6 @@ static inline void net_dim_sample(u16 ev > } > > #define NET_DIM_NEVENTS 64 > -#define BITS_PER_TYPE(type) (sizeof(type) * BITS_PER_BYTE) > #define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) & (BIT_ULL(bits) - 1)) > > static inline void net_dim_calc_stats(struct net_dim_sample *start, > _ >
On Mon, 9 Jul 2018 12:32:51 +0100 Will Deacon <will.deacon@arm.com> wrote: > Hi Andrew, > > On Fri, Jul 06, 2018 at 05:30:49PM -0700, Andrew Morton wrote: > > On Tue, 19 Jun 2018 13:53:08 +0100 Will Deacon <will.deacon@arm.com> wrote: > > > > > In preparation for implementing the asm-generic atomic bitops in terms > > > of atomic_long_*, we need to prevent asm/atomic.h implementations from > > > pulling in linux/bitops.h. A common reason for this include is for the > > > BITS_PER_BYTE definition, so move this and some other BIT() and masking > > > macros into a new header file, linux/bits.h > > > > > > --- a/include/linux/bitops.h > > > +++ b/include/linux/bitops.h > > > @@ -2,29 +2,9 @@ > > > #ifndef _LINUX_BITOPS_H > > > #define _LINUX_BITOPS_H > > > #include <asm/types.h> > > > +#include <linux/bits.h> > > > > > > -#ifdef __KERNEL__ > > > -#define BIT(nr) (1UL << (nr)) > > > -#define BIT_ULL(nr) (1ULL << (nr)) > > > -#define BIT_MASK(nr) (1UL << ((nr) % BITS_PER_LONG)) > > > -#define BIT_WORD(nr) ((nr) / BITS_PER_LONG) > > > -#define BIT_ULL_MASK(nr) (1ULL << ((nr) % BITS_PER_LONG_LONG)) > > > -#define BIT_ULL_WORD(nr) ((nr) / BITS_PER_LONG_LONG) > > > -#define BITS_PER_BYTE 8 > > > #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(long)) > > > -#endif > > > > Why does it leave BITS_TO_LONGS() in place? > > > > That becomes unfortunate with Chris's patch, so I'm moving > > BITS_TO_LONGS() into bits.h. > > The reason I avoided that was because it would pull in the dreaded > kernel.h for DIV_ROUND_UP, and then we're back to circular include hell :( > Well we should comment that so every reader doesn't wonder what I wondered. Refactoring works well. I suppose DIV_ROUND_UP and friends await a new <arithmacros.h>. Also, all these macros no longer `#ifdef __KERNEL__' protection, which wasn't mentioned in the changelog. Deliberate?
Hi again, Andrew, On Mon, Jul 09, 2018 at 03:10:06PM -0700, Andrew Morton wrote: > On Mon, 9 Jul 2018 12:32:51 +0100 Will Deacon <will.deacon@arm.com> wrote: > > On Fri, Jul 06, 2018 at 05:30:49PM -0700, Andrew Morton wrote: > > > On Tue, 19 Jun 2018 13:53:08 +0100 Will Deacon <will.deacon@arm.com> wrote: > > > > > > > In preparation for implementing the asm-generic atomic bitops in terms > > > > of atomic_long_*, we need to prevent asm/atomic.h implementations from > > > > pulling in linux/bitops.h. A common reason for this include is for the > > > > BITS_PER_BYTE definition, so move this and some other BIT() and masking > > > > macros into a new header file, linux/bits.h > > > > > > > > --- a/include/linux/bitops.h > > > > +++ b/include/linux/bitops.h > > > > @@ -2,29 +2,9 @@ > > > > #ifndef _LINUX_BITOPS_H > > > > #define _LINUX_BITOPS_H > > > > #include <asm/types.h> > > > > +#include <linux/bits.h> > > > > > > > > -#ifdef __KERNEL__ > > > > -#define BIT(nr) (1UL << (nr)) > > > > -#define BIT_ULL(nr) (1ULL << (nr)) > > > > -#define BIT_MASK(nr) (1UL << ((nr) % BITS_PER_LONG)) > > > > -#define BIT_WORD(nr) ((nr) / BITS_PER_LONG) > > > > -#define BIT_ULL_MASK(nr) (1ULL << ((nr) % BITS_PER_LONG_LONG)) > > > > -#define BIT_ULL_WORD(nr) ((nr) / BITS_PER_LONG_LONG) > > > > -#define BITS_PER_BYTE 8 > > > > #define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(long)) > > > > -#endif > > > > > > Why does it leave BITS_TO_LONGS() in place? > > > > > > That becomes unfortunate with Chris's patch, so I'm moving > > > BITS_TO_LONGS() into bits.h. > > > > The reason I avoided that was because it would pull in the dreaded > > kernel.h for DIV_ROUND_UP, and then we're back to circular include hell :( > > > > Well we should comment that so every reader doesn't wonder what I > wondered. > > Refactoring works well. I suppose DIV_ROUND_UP and friends await a new > <arithmacros.h>. Yes; pulling apart kernel.h would certainly help solve some of the fragility here. > Also, all these macros no longer `#ifdef __KERNEL__' protection, which > wasn't mentioned in the changelog. Deliberate? Yes, these aren't under uapi/ so I dropped the guards now that they're not needed (my understanding is that they're only there because of the way the uapi split was originally scripted). I can also confirm that the generated user headers match exactly with and without my changes. Will
diff -puN include/linux/bitops.h~bitops-introduce-bits_per_type include/linux/bitops.h --- a/include/linux/bitops.h~bitops-introduce-bits_per_type +++ a/include/linux/bitops.h @@ -11,7 +11,8 @@ #define BIT_ULL_MASK(nr) (1ULL << ((nr) % BITS_PER_LONG_LONG)) #define BIT_ULL_WORD(nr) ((nr) / BITS_PER_LONG_LONG) #define BITS_PER_BYTE 8 -#define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_BYTE * sizeof(long)) +#define BITS_PER_TYPE(type) (sizeof(type) * BITS_PER_BYTE) +#define BITS_TO_LONGS(nr) DIV_ROUND_UP(nr, BITS_PER_TYPE(long)) #endif /* diff -puN include/linux/net_dim.h~bitops-introduce-bits_per_type include/linux/net_dim.h --- a/include/linux/net_dim.h~bitops-introduce-bits_per_type +++ a/include/linux/net_dim.h @@ -363,7 +363,6 @@ static inline void net_dim_sample(u16 ev } #define NET_DIM_NEVENTS 64 -#define BITS_PER_TYPE(type) (sizeof(type) * BITS_PER_BYTE) #define BIT_GAP(bits, end, start) ((((end) - (start)) + BIT_ULL(bits)) & (BIT_ULL(bits) - 1)) static inline void net_dim_calc_stats(struct net_dim_sample *start,